SignalCapabilitySG-0063
Google DeepMind's AI agents formally verified Signal's encryption code down to machine code in four weeks
Working from specifications human experts approved, Gemini 4 Carbon agents wrote about 6.5 million lines of machine-checked Lean proof for libsignal in four weeks. The proofs cover forward secrecy, authentication and post-compromise security for one-to-one sessions. They do not cover group messaging, key generation or post-quantum healing, and they assume the low-level cryptographic routines are correct. When two previously fixed bugs were put back, the system proved the code insecure. DeepMind says no implementation of a protocol this widely deployed had been formally verified before, by any method.
Why it matters
Formal proof rules out whole classes of attack instead of fixing bugs one at a time, and DeepMind says it did in four weeks, with a team of five, work that has historically taken expert teams years. If proofs become cheap, the hard part moves to deciding what to prove. The agents also ran long autonomous campaigns whose output is trusted only because Lean's checker, not the agents, vouches for it.
Sources
Read the reporting
3 sources. Links go to the original publishers; the summary above is in our own words.
- primaryIntroducing AlphaProof Code: a breakthrough in verifying complex software at depthGoogle DeepMind · Nov. 10, 2026deepmind.google/blog/introducing-alphaproof-code/
- researchAlphaProof Code: Verifying Signal's Security at the Machine-Code LevelGoogle DeepMind · Nov. 10, 2026deepmind.google/documents/36/Formal_Verification_of_Signal_4.pdf
- newsGoogle DeepMind's AI agents formally verify Signal's encryption codeTNW · Nov. 12, 2026thenextweb.com/news/google-deepmind-alphaproof-code-signal-encryption-formal-ve…
Signals
More from the news desk
METR found Claude Opus 5.5 could pick out every alignment test from ordinary tasks in its set
Attackers ran AntiFox, an open-source AI pentesting agent forked from ARTEX, against web servers and AWS accounts
METR estimates Anthropic's models are improving about twice as fast as before, and says AI likely drives most of it
Cite and share
Use this signal
Citation
Paperclip Index. “Google DeepMind's AI agents formally verified Signal's encryption code down to machine code in four weeks.” Signal SG-0063. Reported Nov. 10, 2026; updated Nov. 12, 2026. https://paperclipindex.com/signal/SG-0063